Selected AI startups from YC’s Winter ’22 batch – MovieUpdates

Dozens of startups in Y Combinator’s Winter 2022 cohort are doing something that could be described as artificial intelligence. While the term has lost much of its meaning, it is still an important part of the technology landscape, and both using and enabling it is fertile ground for new businesses. Here are 14 notable startups from the latest batch.


Founded: 2022 in San Francisco, California

Through: Jay Chia, Sammy Sidhu

Building: A turnkey data warehouse for image and video. Any recording, organization and processing of image data all on one platform. Also handles queries, easy integration with cloud providers, and smart scheduling to save costs on computing.

Quote: “Currently, users are luging around multiple vendors or difficult-to-manage open-source frameworks. We are a one-stop shop.”

TC Quick Take™: Many companies need some sort of image analysis pipeline and creating a one-stop shop is a valid approach to the market. However, it can be challenging to keep customers who are starting to scale and peer into costs.

Whitelab Genomics

Founded: 2019 in Paris, France

Through: Julien Cottineau, David Delbourgo

Building: An AI-powered platform for discovering and designing genomic therapies. Data science and databases for “scientific, technical, biological, genomic and experimental” datasets. Plus suggestions for experimental design and workflows.

Quote: “We help our customers develop more genomic medicines, faster, in a leaner way and to make those revolutionary medicines more accessible to patients in need.”

TC Quick Take™: Discovering AI drugs is hot, but few have created serious value. Still, of course, it holds promise and specializing in genome therapy is a good idea. How they will do the woollier workflow and design stuff I don’t know.


Founded: 2021 in Mannheim, Germany

Through: Alex Conway, Sascha Lang

Building: A production line control system that registers, digitizes and monitors factories with many workers. Includes real-time alerts, automated reports, and photo-based quality control.

Image Credits: AISupervision

Quote: “We automate what the best supervisors would do if they kept an eye on everything that happens in your factory.”

TC Quick Take™: If it’s anything like what Amazon already does, it’s a short drive to hell for the workers. But if they can make this humane and useful, many factories would see it as a way to integrate multiple tracking tools.

Image Credits: Powerhouse AI

Powerhouse AI

Founded: 2021 in Singapore

Through: Kushal Pillay and Ivo Verhaegh

Building: An automated inventory solution for warehouses. Many labor hours are used to perform this thankless task, so why not automate it? Powerhouse uses regular telephones that count stock using regular photos taken by the staff. It’s part of a wider effort to make warehouses smarter and more efficient.

Quote: “They just take a picture with their phone and our software counts the boxes.”

TC Quick Take™: I thought I heard the presenter say 97% accuracy, but the site says 99%. I don’t do an inventory these days, but hopefully it can be trusted enough that it doesn’t have to be counted by hand.

Strong computing power

Founded: 2021 in Sydney, Australia

Through: Ben Sand

Building: An ML training platform that focuses on ultra-high performance and efficiency. By performing optimizations “around” a model (such as precalculating values, identifying bottlenecks, etc.) they claim to improve training times by orders of magnitude.

Quote: “The future of Cloud Computing, priced by performance, not consumption.”

TC Quick Take™: Frederic looked at it a few weeks ago, and it seems like a nice luxury competitor that plays nicely with existing solutions.

Reality Defender

Founded: 2021 in New York

Through: Gaurav Bharaj, Benjamin Colman and Ali Shahriyari

Building: An easy-to-use tool that allows businesses to scan media for deepfake content. Produces alerts, report cards, and other ways to visualize and take action on fake content.

Quote: “Enable humanity to recognize the truth.”

TC Quick Take™: This will almost certainly be an arms race, but a tool to catch the low-hanging fruit will be necessary to prevent oceans of fake profiles, lightly cloaked spam content and so on. However, their methods can be constantly refined, which may not be cheap or easy.


Founded: 2020 in Canada

Through: Rahul Anand and Rob McEwan

Building: A geospatial and satellite data platform for mining operations that brings together multiple resources and provides a more efficient way to locate and organize mineral collection. Early cases have reduced the turnaround time from a year to a few weeks.

Quote: “We will change mining more in the next five years than in the last 50.”

TC Quick Take™: We’ve seen a lot of movement in the mining area lately, probably a sign that this legacy industry is investing its significant resources in modernization. It’s a promising space to enter now.

variable AI

Founded: 2021 in Pittsburgh

Through: Omar Shams

Building: A machine learning powered polishing engine for Python code and Jupyter notebooks. These notebooks are often cobbled together as experiments with no plan to be production code – but then one is adopted as a popular tool and has to be cleaned up. Mutable AI does this automatically, autocomplete, refactoring and minimizing quickly and easily.

Before (left) and after.

Quote: “Be kind to your teammates and future self.”

TC Quick Take™: Would it be helpful? Probably… I’ve had to use hacked notebooks and wouldn’t want to do it full time. Is there a big market for it? Hard to say. The company seems to be born out of the founder’s own frustrations on the fringe of ML research, so maybe others like him are looking for this.


Founded: 2020 in Berlin, Germany

Through: Fabio and Marcel Schmidberger, Erik Ziegler

Building: Voice-controlled filling of forms for healthcare workers. These people spend hours doing paperwork and it’s a sea of ​​fields to browse and type. With Voize, they can speak directly into a smartphone and structured records are created from it.

Quote: “The digital voice assistant for non-desk workers.”

TC Quick Take™: Speaking medical information out loud as you walk around seems like a big no-no… I feel like there’s a reason it was done the way it was done. But streamlining the medical records definitely needs to be done.


Founded: 2021 in San Francisco

Through: Hahnbee Lee and Han Wang

Building: An automatic code documentation engine that reads your code and securely inserts comments that put it into context. Coders don’t always have time to do that and it makes checking and reusing code difficult – so having a tool that does this for them (at least to some degree) can save time and effort.

Quote: “Everyone suffers from documentation debts.” Also regarding co-founder Han Wang: “Knows more than 100 cow jokes.” We’ll see.

TC Quick Take™: The proof is in the pudding for this one, but I bet a lot of companies would be happy to say that every line of code is explained. “Mintlise it before you send it to me next time!”


Founded: 2016 in Helsinki, Finland

Through: Hannes Heikinheimo and Otto Soderlund

Building: On-device voice recognition for apps that don’t want to call cloud service. It is preferable to run each process locally, but until recently there was too much of a trade-off between accuracy and speed. If Speechly can create fast and accurate speech recognition that easily fills forms or retrieves relevant information, it’s one less online API call to make.

Quote: “Reduce cloud costs, improve privacy, and deliver zero-latency experiences by running Speechly directly on the end user’s device.”

TC Quick Take™: Others certainly aspire to this (including monsters like Google and Amazon), but having an independent and functional solution can be very valuable for app makers who want to keep their code as local as possible.


Founded: 2021 in San Francisco

Through: Christian Lau and Vaikkunth Mugunthan

Building: A plug-and-play federated learning platform for AI models in privacy-critical industries. Federated learning is an established technique and has already generated a lot of value, but it is not the easiest to deploy. DynamoFL gets into training workflows that use private data and makes or keeps it private.

Quote: “In three minutes we can connect our federated learning module to your existing ML workflow.”

TC Quick Take™: If it were easier to use federated learning, more niche sectors in healthcare, finance and so on would likely adopt ML models. Whether they can maintain the general ease of use that other ML tools have cultivated is a question mark.


Founded: 2019 in Houston, Texas

Through: Alex Arevalos, Drew Hendricks, Hannah McKenney

Building: AI-powered home urinalysis using a bathroom spectrometer. Using a mostly lab-based device for identifying and detecting substances, Starling tries to catch things like bladder infections and other problems early, or perform home monitoring that normally requires a urine sample to be brought in.

Image Credits: Starling

Quote: “What information did you just flush down the toilet?”

TC Quick Take™: Depending on how easy it is to use, this could be a new standard tool for many medical preventative cases. I can see it everywhere in two years.

Sieve data

Founded: 2021 in Berkeley, California

Through: Abhinav Ayalur and Mokshith Voodarla

Building: A dead simple video analytics platform that scans incoming video for things like people, objects, and situations and returns that metadata quickly. Enables search, storage, moderation – everything you need to understand video to do.

Quote: “You never have to train a model or manage a dataset. Just press video and get results.”

TC Quick Take™: I just wrote a company called Twelve Labs that does something like this and it seems like a promising niche. It is difficult to pick winners at this stage, it will depend a lot on new cases of video use.

Read more about YC Demo Day at MovieUpdates

Show Love ❤️